Personalized assessment of bone health

1. A computer-implemented method for personalized assessment of a subject's bone health, the method comprising: extracting a plurality of features of interest from non-invasive subject data, medical images of the subject, and subject-specific bone turnover marker values; using a surrogate model and the plurality of features of interest to predict one or more subject-specific measures of interest related to bone health; and generating a visualization of the one or more subject-specific measures of interest related to bone health; wherein the surrogate model is trained by a process comprising: retrieving training data comprising one or more of (i) a plurality of bone anatomical models and (ii) a plurality of in-vitro models from a database; performing FEM-based computations using the plurality of bone anatomical models or stress-experiments using the plurality of in-vitro models to yield FEM results; extracting one or more measures of interest from the FEM results; extracting a plurality of geometric features from the plurality of bone anatomical models; and training the surrogate model to predict the one or more measures of interest based on the plurality of geometric features using a machine learning algorithm.

2. The method of claim 1 , wherein the measures of interest comprise one or more of stress and stress strain.

3. The method of claim 1 , wherein at least a portion of the training data comprises synthetic data.

4. The method of claim 3 , wherein the synthetic data is generated by: generating one or more baseline models; randomly or systematically perturbing the baseline models to obtain a plurality of synthetic models comprising one or more of (i) synthetic bone anatomical models and (ii) synthetic in-vitro models.

5. The method of claim 4 , wherein the baseline models are subject-specific anatomical models.

6. The method of claim 3 , wherein the synthetic data comprises one or more of (i) synthetic bone anatomical models and (ii) synthetic in-vitro models generated according to a set of rules using one or more randomly or systematically perturbed parameter values.

7. The method of claim 1 , further comprising: associating each of the one or more subject-specific measures of interest with a point on a subject image; and displaying the subject image; and in response to receive a user selection of a selected point on the subject image, displaying a particular subject-specific measures of interest corresponding to the selected point.

8. The method of claim 1 , further comprising: associating each of the one or more subject-specific measures of interest with a point on a subject image; and displaying the subject image color coded based on values of the subject-specific measures of interest.

9. A parallel processing computing system comprising: a host computer configured to extract a plurality of features of interest from non-invasive subject data, medical images of a subject, and subject-specific bone turnover marker values; and a device computer configured to use a surrogate model and the plurality of features of interest to predict one or more subject-specific measures of interest related to bone health, wherein the surrogate model is trained by a process comprising: retrieving training data comprising one or more of (i) a plurality of bone anatomical models and (ii) a plurality of in-vitro models from a database; performing FEM-based computations using the plurality of bone anatomical models or stress-experiments using the plurality of in-vitro models to yield FEM results; extracting one or more measures of interest from the FEM results; extracting a plurality of geometric features from the plurality of bone anatomical models; and training the surrogate model to predict the one or more measures of interest based on the plurality of geometric features using a machine learning algorithm.